Morse Theory

Last modified date: <%+ tp.file.last_modified_date() %>



Morse Theory

The standard procedure:

  • Show that D is a Fredholm operator
  • Show that D is surjective, so IndD=dimkerD
  • Show a moduli space is the intersection of some section s of a bundle with the zero section.
  • Show that this intersection is transverse, i.e. Ds is surjective.
    • Vary the Riemannian metric and use a second category theorem to get the Morse-Smale condition.
  • Apply the infinite-dimensional inverse function theorem
  • Show that a Frechet manifold is in fact a Banach manifold and apply a version of Sard’s Theorem

attachments/Pasted%20image%2020220401094806.png attachments/Pasted%20image%2020220401094918.png

Motivations

Results

Theorem: Every compact smooth manifold admits a Morse function.

Theorem: Morse function are generic: given any smooth function f:XY, there is an arbitrarily small perturbation of f that is Morse.

See Morse lemma

Theorem 3: If (W;M0,M1)I is Morse with no critical points then WDiffI×M0

Theorem: If X is closed and admits a Morse function with exactly 2 critical points, X is homeomorphic to Sn.

Possibly used in Milnor’s Unsorted/parahoric 7-sphere (show a diffeomorphism invariant differs but admits such a Morse function)

Theorem: M is homotopy equivalent to a CW complex with one cell of dimension k for each critical point of f of index of a Morse function k.

attachments/Pasted image 20210501235532.png

attachments/Pasted%20image%2020220422101431.png attachments/Pasted%20image%2020220422101514.png attachments/Pasted%20image%2020220422101532.png attachments/Pasted%20image%2020220422101540.png

Gradients

attachments/Pasted%20image%2020220422121526.png attachments/Pasted%20image%2020220422121549.png attachments/Pasted%20image%2020220422122055.png # Energy

attachments/Pasted%20image%2020220422121652.png attachments/Pasted%20image%2020220422121714.png

Critical points

Idea: the number of linearly independent direction you can move for which the function decreases.

attachments/Pasted%20image%2020220422120052.png attachments/Pasted%20image%2020220422120201.png

Morse chain complex

attachments/Pasted%20image%2020220422120306.png attachments/Pasted%20image%2020220422120319.png

Morse inequalities

attachments/Pasted%20image%2020220422120836.png

Broken trajectories

attachments/Pasted%20image%2020220422121840.png

Moduli space of flow lines

attachments/Pasted%20image%2020220422122307.png attachments/Pasted%20image%2020220422122322.png attachments/Pasted%20image%2020220422122513.png

Zero set of a section

attachments/Pasted%20image%2020220422123136.png

See vertical and horizontal subspace

attachments/Pasted%20image%2020220422123254.png

Sard

attachments/Pasted%20image%2020220422123700.png

Pictures

attachments/Pasted%20image%2020220422120347.png

Examples

attachments/Pasted%20image%2020220422120415.png

attachments/Pasted%20image%2020220422120442.png

attachments/Pasted%20image%2020220422120518.png

attachments/Pasted%20image%2020220422120603.png attachments/Pasted%20image%2020220422120616.png

attachments/Pasted%20image%2020220422120637.png attachments/Pasted%20image%2020220422120712.png attachments/Pasted%20image%2020220422120725.png

Notes

Dave’s Videos

  • Historic note: Morse wanted to know not the diffeomorphism type of M, but rather the homotopy type.

  • Definition: critical values and critical points

  • Definition: critical point

  • Theorem (Smale, h-cobordism theorem

    • If Xn is a smooth cobordism.
  • Corollary (High-Dimensional Unsorted/Poincare conjectures

    • If Xn1,Xn2DiffSn, then there exists an h-cobordism between them.
    • Proof: use algebraic topology to eliminate (cancel) critical points.
  • Definition: index of a Morse function

    • Look at coordinate-free def?
    • Standard form at critical points
    • Alternatively: Hessian is non-singular at every critical point.
    • f1Y)=X
  • Definition: Stable and generic

  • Definition: cobordism

    • Example: (pair of pants)
    • Category: Objects are manifolds, morphisms are cobordisms between them
  • Consequence of theorem 3: M0DiffM1 is a diffeomorphism, useful to show two things are diffeomorphic, used in higher-dimensional Poincare.

    • Recall that this is proved by constructing a vector field V on W, then using a diffeomorphism ϕ:I×M0W by flowing along V.
    • Can we do gradient flow in the presence of a metric? #todo/questions

Intro Video

https://www.youtube.com/watch?v=78OMJ8JKDqI

Morse theory: handles nice singularities. Can have worse ones, covered by [dynamical systems](catastrophe theory](dynamical%20systems](catastrophe%20theory) (dynamical systems).

Importance of CW complexes: triangulation of surfaces.

See Morse lemma

Morse Theorem 1: If there are no critical points, MAMB.

attachments/Pasted image 20210501235559.png

Stable vs unstable manifolds:

attachments/Pasted image 20210501235734.png

Consider height function on torus. Circles are index 0 critical points, triangle is index 1.

attachments/Pasted image 20210501235700.png

Cancellation:

attachments/Pasted image 20210501235757.png

attachments/Pasted image 20210501235820.png

Can use persistent homology to measure “importance” of critical points.

Unsorted

https://youtu.be/mIUi1zIUQJw?t=42

attachments/Pasted image 20210501235429.png

attachments/Pasted image 20210501235456.png

  • Diffeomorphism type depends on isotopy classes of attaching maps.

See handle decomposition

More Notes

Historic note: Morse wanted to know not the diffeomorphism type of M, but rather the homotopy type. - Theorem (Smale, h-cobordism) - If Xn is a smooth cobordism, n6, π1(X)=0, and X “looks like” a product in algebraic topology, then X is a product cobordism. - Corollary (High-Dim Poincare) - If Xn1,Xn2DiffSn, then there exists an h-cobordism between them. - Proof: use algebraic topology to eliminate (cancel) critical points. - Theorem: Every compact manifold has a Morse function. - Theorem: Morse functions are generic (given any smooth function f:XY, there’s an arbitrarily small perturbation of f that is Morse). - Theorem (Morse Lemma): If pRn is a critical point of f:RnR such that the Hessian Hf(p) is a non-degenerate bilinear form, then f is locally Morse (standard form). - Theorem: If (W;M0,M1)I is Morse with no critical points then WDiffI×M0 - Consequence: M0DiffM1 is a diffeomorphism, useful to show two things are diffeomorphic, used in higher-dimensional Poincare. - Theorem: If X is closed and admits a Morse function with exactly 2 critical points, X is homeomorphic to Sn. - Possibly used in Milnor’s exotic 7-sphere (show a diffeomorphism invariant differs but admits such a Morse function) - Diffeomorphism type depends on isotopy classes of attaching maps.

Morse Theory

Goal: handlebody decomposition, or for the purposes of the above theorems, retractions onto a CW complex. Decomposing a cobordism into a sequence of elementary cobordisms (admit a Morse function with a single critical point).

Fact: since ϕ is Morse, M2n can be retracted onto a complex of dimension dn, since all critical points will have index n.

Note: this immediately implies the Lefschetz Hyperplane theorem for affine manifolds N, i.e. that they are entirely determined by the homology and homotopy of NH for any hyperplane. Very strong!

Setting up notation/definitions:

  • V will be a smooth n-manifold
  • W an n-dimensional cobordism
  • ϕ:VR a smooth function
  • p a critical point of ϕ (i.e. the derivative dpϕ vanishes)
  • Hpϕ=(2ϕx2ix2j) the Hessian matrix
  • \nullϕ(p) the nullity of ϕ at p is dimkerHp, regarding Hpϕ as a symmetric bilinear form on TpV
  • p is nondegenerate iff \nullϕ(p)=0.
  • The Morse index at p is the dimension of the maximal subspace on which the associated quadratic form Hpϕ is negative definite.
Theorem (Morse Lemma)

Near a nondegenerate critical point p of ϕ of index k there exists a smooth coordinate chart U and coordinates xRn such that ϕ has the form ϕ(x)=ϕ(p)+xtAx where A=diag(1,,1,1,1).

Toward a generalization, we can also write R=Rk×Rnk and ϕ(x1,x2)=ϕ(p)

Lemma (The nondegenerate directions can be split off)
If \null_\phi(p) = \ell then we can instead write {\mathbf{R}}= {\mathbf{R}}^{n-k-\ell} \times{\mathbf{R}}^k \times{\mathbf{R}}^\ell and \begin{align*} \phi(\mathbf{x}, \mathbf{y}, \mathbf{z}) = {\left\lVert {\mathbf{x}} \right\rVert}^2 - {\left\lVert {\mathbf{y}} \right\rVert}^2 + \psi(\mathbf{z}) \end{align*} where \psi: {\mathbf{R}}^\ell \to {\mathbf{R}} is some smooth function.
Definition
A degenerate critical point is embryonic iff \null_\phi(p) = 1 and writing L = \ker H_p\phi = \mathop{\mathrm{span}}_{\mathbf{R}}{\mathbf{v}}, the third directional derivative D^3_{\mathbf{v}}\phi (?) is nonzero.

We now consider homotopies of Morse functions \phi: I \times V \to {\mathbf{R}}, where we can partially apply the I factor to get a 1-parameter family \left\{{\phi_t {~\mathrel{\Big\vert}~}t\in I}\right\}.

Definition
A homotopy \Phi: V\times I \to {\mathbf{R}} of Morse functions has a birth-death type critical point at p at t=t_0 iff p is embryonic for \phi_0 and (t_0, p) is a nondegenerate critical point of \Phi.

Recall what a Cerf diagram/profile is – I don’t

Theorem (Whitney)

In three parts:

  • Near an embryonic critical point p of \phi of index k there exist coordinate (\mathbf{x}, \mathbf{y}, z) \in {\mathbf{R}}^{n-k-1} \oplus {\mathbf{R}}^{k} \oplus {\mathbf{R}} such that \phi has the form \begin{align*} \phi(\mathbf{x}, \mathbf{y}, z) = \phi(p) + {\left\lVert {\mathbf{x}} \right\rVert}^2 - {\left\lVert {\mathbf{y}} \right\rVert}^2 + z^3 \end{align*}

  • If p is birth-death type for \Phi of index k, then up to conjugating \phi_t by a (uniform in t) family of diffeomorphisms, each \phi_t is of the form \begin{align*} \phi(\mathbf{x}, \mathbf{y}, z) = \phi(p) + {\left\lVert {\mathbf{x}} \right\rVert}^2 - {\left\lVert {\mathbf{y}} \right\rVert}^2 + z^3 \pm tz \end{align*}

  • Any two homotopies \Phi, \Phi' with points (p, 0) and (p', 0) with the same index and Cerf profile differ only by a diffeomorphism, i.e. there is a family of diffeomorphisms h_t such that \phi'_t \circ h_t = \phi_t for every t.

  • A generic \Phi has only nondegenerate and birth-death type singularities.

Definition
A singularity is birth type if the sign on t is positive, and death type if negative.
Fact
Embryonic critical points are isolated, near a birth-type singularity two nondegenerate critical points of indices k, k-1 emerge, and near a death type they merge and disappear.

Pretty vague – I know there are pictures here that make this more obvious, but I couldn’t find much.

Definition
A cobordism is a triple (W; M_+, M_-) where W is an oriented compact smooth manifold with cooriented boundary {{\partial}}W = M_+ {\textstyle\coprod}M_- = {{\partial}}_- W {\textstyle\coprod}{{\partial}}_+ W, where the coorientation is provided by the inward (resp. outward) normal vector field (???). We’ll usually just denote this as W.
Definition

A Lyapunov cobordism is a triple (W, \phi , X) where

  • W is a usual cobordism,
  • \phi: W\to {\mathbf{R}} is a smooth functional that is constant and has no critical points when restricted to {{\partial}}W,
  • X is a gradient-like vector field for \phi which points inward along {{\partial}}_- W and outward along {{\partial}}_+ W.
Definition
Such a cobordism is elementary iff there exist no X{\hbox{-}}trajectories between distinct critical points of \phi.
Theorem (Smale, h-cobordism)
Let W be a cobordism of dimension W\geq 6 such that W, {{\partial}}_{\pm}W are simply connected, and H_*(W, {{\partial}}_- W; {\mathbf{Z}}) = 0. Then W admits a Morse function without critical points which is constant on {{\partial}}_\pm W.

In particular, W \cong I\times M is diffeomorphic to the trivial product cobordism, and M\cong N are diffeomorphic.

Proof (Sketch)

Goal: find a handle decomposition with no handles, then integrate along the gradient vector field of a Morse function \phi to get a diffeomorphism.

  • Find a Morse function and induce a handle decomposition
  • Rearrange handles so that lower index handles are attached first
  • Define a chain complex as free {\mathbf{Z}}{\hbox{-}}module on handles with boundary given in terms of intersection numbers of attaching spheres k and belt k-1 spheres
  • Find k{\hbox{-}}handles, create a pair of k+1, k+2 handles such that the k+1 handle cancels/fills in the k{\hbox{-}}handle (not sure why the k+2 is needed here)
  • End up with nothing but an n{\hbox{-}}handle and an n-1{\hbox{-}}handle – turn “upside down” and repeat process with -\phi to remove them.

Proof (Sketch)

  • Pick \phi: W\to {\mathbf{R}} Morse such that {{\partial}}_\pm W are regular level sets.
  • Make \phi self-intersecting (uses a transversality argument)
  • Partition manifold into regular level sets L_k \coloneqq\phi^{-1}(k - \frac 1 2) for each k\in {\mathbb{N}}.
  • Letting \left\{{p_i}\right\} be the critical points in L_k and \left\{{q_j}\right\} the critical points in L_{k-1}, form the matrix A of intersection numbers S_{p_i}^- \smile S_{q_j}^+ between the stable sphere of p_i and the unstable sphere of q_j.
  • Goal: since homology can be read off SNF(A), and we know H_* = 0 here, we try to reduce A to SNF with geometric operations
    • Handle slides: Add row j to row i by moving p_i to L_{k+j}, ~j\geq 1, deform X to produce a trajectory p_j \to p_i, then “the stable manifold of p_i slides over the stable manifold of p_j” (?) replacing [S_i^-] with [S_i^-] + [S_j^-] in homology.
    • This makes A = [I, 0; 0, 0] a block matrix with the identity in the top-left.
    • Handle Cancellation: Take two transverse intersection points z_+, z_- with local intersection indices 1, -1, connect via two paths: one in S_i^-, one in S_j^+. This yields a map S^1 \hookrightarrow L_k, use the Whitney trick to fill with an embedded disc \Delta, then push S_i^- over \Delta eliminates z_\pm.
    • This leaves a collection S_i^-, S_i^+ for i=1,\cdots, r intersecting in a single point z_0, then (lemma) there are unique trajectories q_i \to p_i for each I and thus they can be eliminated.
  • Do this in L_k; we now have a Morse function with no critical points except possibly of index 0, 1, n-1, or n.
  • Use “Smale’s trick”: trades in an index k critical point for one of index k+1 and one of index k+2, such that k, k+1 cancel. Trade index 1 for index 2, 3 and cancel index 3 as before.
  • Eliminate 0, n with a lemma (unclear)
Links to this page
#projects/notes/reading #geomtop #resources/notes #resources/videos #todo/questions